Recent Developments in Low-Level Lead Exposure and Intellectual Impairment in Children

Karin Koller; Terry Brown; Anne Spurgeon; Len Levy


Environ Health Perspect. 2004;112(9) 

In This Article

Prospective (Longitudinal) Studies

The evidence for lowered intellectual and cognitive ability in children exposed to lead comes largely from prospective epidemiologic studies of cohorts in Boston, Massachusetts, USA; Cincinnati and Cleveland, Ohio, USA; Port Pirie and Sydney, Australia; and Yugoslavia (Factor-Litvak et al. 1999). A number of these studies are still ongoing. The main focus of current debate centers on the difficulties of adjusting for confounders (covariables), which include socioeconomic status (SES), home environment, and genetic factors. SES is measured in a number of ways that generally involve an index derived from data on household income, parents' education, employment status, occupation, and home ownership. Home environment is frequently measured using the Home Observation for Measurement of the Environment Inventory (HOME) index. This reflects the quality and quantity of emotional and cognitive stimulation and support in the home environment. The total score is the sum of a number of items, each scored as present (1) or absent (0), in various categories: parental responsivity, acceptance of child, organization of home environment, provision of play materials, parental involvement with child, and variety of stimulation.

Longitudinal studies have many advantages over cross-sectional studies: a) the time sequence of events can be assessed, b) they can provide information on a wide range of outcomes, and c) there is reduced recall and selection bias compared with case-control studies. Children's intellectual capacities change with time, and therefore age-specific tests must be used (i.e., there is no single psychometric test that can cover the entire age range of interest). Unfortunately, in the five ongoing lead/IQ studies identified, a variety of cognitive test instruments were used, even for children of the same age, and no two studies adjusted for the same covariables. It is therefore not possible to directly compare results between these studies.

The Yugoslavia Prospective Lead Study was initiated in 1985. Pregnant women (n = 1,502) living in two towns in Yugoslavia were identified as having differing lead exposures. One town is on the site of a lead smelter, whereas the other (control) town lies 25 miles to the south. Maternal blood lead was measured at midpregnancy and at delivery, and child blood leads were determined at subsequent 6-month intervals. The report by Wasserman and colleagues (2000) includes all children (n = 390) having at least one assessment of intellectual functioning at 3, 4, 5, or 7 years of age with complete data on all covariates. Three normed and age-specific tests of cognitive function were used. This review is a reanalysis of data given in the authors' full 1999 report, when the study was in its fourteenth year (Factor-Litvak et al. 1999), and examines whether there are critical time periods for the effects of lead exposure on IQ. Data analysis was performed after grouping observations into three exposure-change categories. An increase in postnatal blood lead was defined by a change of 50% or more relative to prenatal levels, with the postnatal period divided into early (0-2 years of age) or late (2-7 years of age). Average prenatal blood lead levels were 10 µg/dL (range 3-30 µg/dL) and average postnatal levels at 2-7 years of age were 17.4 µg/dL (range 6.6-49 µg/dL). The wide ranges are a reflection of pooling data from children living in two towns with very different exposure levels.

Both prenatal (p < 0.001) and postnatal (p < 0.05) exposure were independently and significantly negatively correlated with IQ, and no critical period of vulnerability was found. A 50% rise in prenatal blood lead levels was associated with a 1.07-point loss in IQ score [95% confidence interval (CI), 0.6-1.53], whereas a 50% increase in postnatal blood lead relative to prenatal levels was associated with a 2.82-point IQ loss (95% CI, 0.52-4.91). Because the analyses first controlled for prenatal blood lead, the postnatal change measure indicated a substantial change in exposure and was not a reflection of whether the mean postnatal blood lead was high or low. Covariates included in the regression analysis were quality of the home (HOME score); maternal age, intelligence, education and ethnicity; birth weight; and sex. Together these accounted for approximately 50% of the variance in IQ at 7 years of age; lifetime lead exposure accounted for 4.2% of the variance (Factor-Litvak et al. 1999).

The report by Schnaas and colleagues (2000) forms part of the Mexico City Prospective Lead Study of 436 children born after uncomplicated pregnancy. Intellectual function was measured using the McCarthy Scale, which provides a general index of intellectual ability (General Cognitive Index, GCI); subtests measure both cognitive and motor function. Complete data were obtained for 112 children followed at 6-month intervals between 3 and 5 years of age. Prenatal blood lead measures were recorded at intervals during pregnancy, at delivery, and in cord blood. Average postnatal blood lead levels were calculated for three time periods: 6-18 months, 24-36 months, and 42-54 months. Geometric mean blood lead concentrations were approximately 10 µg/dL during the study period. Covariates used in regression models were maternal IQ, child's sex, Apgar score at 5 min, birth weight, birth rank order, maternal educational level and IQ, and family SES (no details given). The authors did not include HOME scores, claiming that HOME scores are highly correlated with maternal IQ and it was therefore sufficient to include only maternal IQ in the model. The article is methodologically very complex, with interactions measured between many variables. The central finding is that prenatal log-transformed blood lead levels are not associated with intellectual function, either within or between subjects, whereas postnatal lead levels were significantly correlated with intellectual function. The strength of the association between mean blood lead (6-18 months) and GCI increases with age up to 4 years of age, after which it becomes less strong and decreases toward zero. This study is one of only a few that attempts to examine in detail the temporal pattern of the association of lead levels and intellectual function.

Tong et al. (2000) provide an update on the 375 children born in the lead-smelting city of Port Pirie, South Australia, who have been followed from birth and had reached 11-13 years of age at the time of the study. Previous studies of this cohort had shown that blood lead concentration was negatively associated with cognitive performance, with girls more sensitive to the effects of lead at 2, 4, and 7 years of age (Baghurst et al. 1992; Tong et al. 1996, 1998). Geometric mean blood lead levels in this cohort increased from 8.3 µg/dL at birth to 21.2 µg/dL at 2 years of age and decreased to 7.9 µg/dL at 11-13 years of age. This study explores whether there is any effect modification between lead exposure and key sociodemographic factors on IQ [measured using the Wechsler Intelligence Scales for Children-Revised (WISC-R) instrument] at 11-13 years of age. A large number of covariates were measured. Sociodemographic factors included sex, maternal IQ, HOME scores, and SES (estimated by Daniel's scale of prestige of occupations in Australia). The cohort was divided into three groups on the basis of lifetime average blood lead levels, with the lowest group < 12 µg/dL and the highest group > 17 µg/dL. The effect of sex became statistically insignificant at 11-13 years of age, and the authors speculate that this may have been due to attrition in numbers. (The original cohort comprised 723 children.) The impact of lead on IQ was more marked in children with lower SES, although this became nonsignificant after adjusting for covariates. The high-SES children performed significantly better in arithmetic and vocabulary WISC-R subscales than children from poor SES backgrounds. Adjusted regression coefficients showed boys lost 2.6 IQ points (95% CI, 2.9 to -8.0), whereas girls lost 7.4 IQ points (95% CI, -1.7 to -13.1) for each 2.7-fold increase in lifetime average blood lead level.

The study by Emory and colleagues (2003) examined 79 mother-infant pairs who represented an independent sample drawn from a larger population of more than 500 subjects in an ongoing study of lead exposure. Mothers came from an urban cohort of low-SES African Americans in Atlanta, Georgia, USA, and their infants were included in the study if they were born after uneventful pregnancies. Maternal blood lead was measured at 6 months' gestation and before delivery and compared with infant memory at 7 months, assessed by the Fagan preferential-looking test. This study was noteworthy for its use of more sensitive calibration standards and continual verification reference samples to increase confidence in measuring very low blood lead levels (< 5 µg/dL). Mean maternal blood lead was 0.72 ± 0.86 µg/dL. Umbilical blood lead was measured, but no data were given in the article. Infant Fagan scores were classified as low, medium, or high risk of later mental retardation. Significant negative correlations between maternal blood lead and subsequent infant Fagan ratings were reported. These differences were not related to gestational age, birth weight, or age at testing nor were they related to mother's education, although it was not stated how this was measured. Overall, these findings should be treated cautiously because of the small numbers in the low- and high-risk groupings, and the lack of detailed information on confounders. The authors acknowledge that their results require replication. This is an ongoing study, and it will be of interest to follow future publications.

The recent article by Canfield and colleagues (2003) is part of this continuing evidence base and relates to a cohort of 240 children born between July 1994 and January 1995, living in Rochester, New York, USA, and enrolled in the Rochester Longitudinal Lead Study. This study population is a nested cohort within a larger group of children and their families who took part in a 24-month randomized dust-control trial published in 1999 (Lanphear et al. 1999). The article by Canfield and colleagues reports on the results of blood lead concentrations measured at 6, 12, 18, 24, 36, 48, and 60 months of age, and IQ determined at 3 and 5 years of age using the Stanford-Binet Intelligence Scale. The relations between blood lead levels and IQ were estimated with a variety of models, with adjustments for nine prespecified covariables: child's sex, birth weight, iron status, and home environment (HOME scale conducted by face-to-face interview and direct observation within the home) (Canfield R, personal communication); mother's IQ, years of education, race, and tobacco use during pregnancy; and household income. Adjusting for number of siblings and birth order did not alter the model estimates or significance levels, and these covariates were therefore not included in the secondary analyses (Canfield R, personal communication).

The study reports a significant negative association (p = 0.004) between blood lead levels and IQ, with a 0.46-point decrease in IQ for each microgram per deciliter increase in lifetime average blood lead concentration (lifetime being equivalent to the child's total exposure over 3 or 5 years). For the subsample of children whose maximal blood lead level remained below 10 µg/dL over the 5 years, the IQ loss associated with a given change in blood lead level was greater. In these 101 children, the study indicates a loss of 0.74 IQ points for each microgram per deciliter increase in lifetime blood concentration. The authors suggest a nonlinear relationship between children's IQ scores and their blood lead concentration, with larger associations at lower lead concentrations. The importance of this study is that it extends the association of blood lead concentrations and intellectual impairment to concentrations below the current level of concern, which stands at 10 µg/dL (0.483 µmol/L), and implies that there is no safety margin at existing exposures.

To fully evaluate the results of the Canfield study, the experiences of their (nested) study cohort within the original dust-control trial must be considered. The Canfield cohort comprised 240 children from a larger group of 276 children and their families taking part in the dust-control trial. Families were eligible for the dust-control trial if they lived in the city of Rochester and had a child 5-7 months of age at the time of the baseline visit. Participants were identified using sequential lists of live births from three urban hospitals, and families were recruited by telephone. Families who agreed to participate were visited by a study team who carried out a baseline interview and collected a venous blood sample from the child. In addition, an experienced technician collected and analyzed dust samples at various indoor locations and measured lead content of painted surfaces inside and outside the home. This original cohort was randomly divided into an intervention group (n = 140) and a control group (n = 135). Families in the intervention group received cleaning equipment and up to eight visits by a dust-control advisor, although the length of time between visits was not specified. All families continued to be visited by the study team at 6-month intervals for blood sampling and environmental lead measurements by a technician (blinded to intervention status). In addition, at each of these home visits an interviewer (also blinded) conducted a face-to-face interview to identify, among other things, the type and frequency of cleaning and the last time cleaning was performed.

Over the 18-month dust-control study period, there was a 2.6-fold increase in blood lead levels in all children, but no difference in blood lead levels by intervention status (geometric mean level 2.85 µg/dL at 6 months, 7.55 µg/dL at 24 months). House dust lead levels declined sharply in both the intervention and control groups. Six months after the first baseline visit, dust lead levels in interior window sills and on floors had decreased by approximately 50% and continued to decline at a slower rate over the following year. The authors recognized several limitations of the study, including sampling the same location in each house, that is, the act of sampling itself may have introduced an artificial decline in dust lead levels. Another possibility was that the act of sampling altered the cleaning behavior of the control group families (the Hawthorne effect). To examine whether the regular visits and dust sampling introduced such an effect, birth certificate data were used to construct a matched negative-control group of 236 children. Children were matched by race, month of birth, and poverty level (measured by census block group characteristics). At 24 months of age the geometric mean blood lead levels were 7.3 µg/dL (95% CI, 6.6-8.2) in the intervention group, 7.8 µg/dL (95% CI, 6.9-8.7) in the control group, and 7.3 µg/dL ± 2.2 µg/dL (CI not given) in the matched negative controls. No Hawthorne effect was apparent.

If it is assumed that the matched negative-control group lived in homes with dust lead levels equivalent to those found in the study cohort before any interventions (baseline values), house dust lead levels do not appear to correlate with blood lead levels in the study children. This is not discussed in the 1999 article by Lanphear et al. (1999) but highlights the difficulties of accurately measuring lead levels in the personal environment of young children. It is interesting that data on house dust lead levels were not included in a follow-up report on dust control and blood lead levels when these children attained 48 months of age (Lanphear et al. 2000).

Canfield et al. (2003) analyze the original intervention and the control children as a single population, and it is pertinent to ask whether any of the interventions in the original dust-control study may have affected blood lead or IQ levels in their (nested) study group. Although there was no significant difference in blood lead between the two groups from 12-24 months of age, the intervention group mean blood levels were 5-7% lower than the controls. If dust control had altered the variability in blood lead levels, this could have affected the power of the study to look at associations with IQ. The second question to ask is, "Could the up to eight extra visits from the dust-control advisor have resulted in a more stimulating learning environment for children in the intervention group compared with controls?" These families had extra visits by one of two randomly assigned advisors, with the provision and replenishment of cleaning equipment and supplies (brooms, dustpans, sponge mops, buckets, gloves, and detergents). Mean IQs of study children and their mothers were below the national average, commensurate with sample demographics. The IQs of the children were normally distributed, whereas IQs of the mothers were slightly skewed because of a larger than expected number of observations in the 70-75 range (Canfield R, personal communication). Because of low sample numbers no significance should be attached to this finding. If these interventions enhanced the cognitive development of the children, they would have resulted in a shift in the relationship between IQ and blood lead levels, with a higher IQ for a given lifetime blood lead level in half of the study children. The effect of this would be to reduce the overall contribution of lead exposure to intellectual impairment. Therefore, in the context of the results of the Canfield study, the previous experiences of their study cohort might have biased the study findings toward the null, that is, attenuated the association between blood lead level and IQ.

Nonlinear mixed models were analyzed using the full range of blood lead values. Figure 1 illustrates the unadjusted lifetime average blood lead and IQ values. The authors state that the cluster of 10 children with low blood lead levels and high IQ "were not unduly influential in the statistical models," and regression diagnostics did not identify any outliers in the data. Secondary analyses (using lifetime average blood lead levels) were carried out on the basis of observations with IQ scores < 110 (Canfield R, personal communication). For the full model this eliminated the 16 highest IQ scores. The overall linear regression coefficient for the remaining subgroup was -0.44 (p = 0.005), which is not significantly different from the coefficient of -0.46 for all 172 children. For the group of children with a peak blood lead concentration < 10 µg/dL, 15 observations were eliminated by using the IQ < 110 cutoff. In this case the linear regression coefficient was -1.07 (p = 0.038), which again is not significantly different from the coefficient of -1.37 for all children with a peak blood lead < 10 µg/dL. However, after eliminating these observations, the p-value decreased from 0.05 to 0.08 in the quadratic model.

Intelligence quotient as a function of lifetime average blood lead concentration. Data were modified from Canfield et al. (2003).

It would be of interest to know a) if removal of data from the 140 children in the original dust-control intervention cohort alters the semiparametric analysis relationship given in Figure 1, and b) to which group the cluster of 10 children with high IQs (> 115) and low blood lead levels (< 5 µg/dL) were assigned in the original study. In addition, data from the nine children with the highest blood lead levels may have had a disproportionate influence on the final slope of the curve compared with subjects clustered around the average blood lead level, and further information on these nine children would also be of interest.

Cognitive function was assessed using an abbreviated Stanford-Binet Intelligence Scale (version IV) at 3 and 5 years of age, with a different examiner administering the test at each age. Results are expressed as the composite score. However, this test may not have been the most accurate measure of IQ for this cohort. The Stanford-Binet is heavily weighted on verbal skills and has been superseded by the Wechsler scales for this reason. Anyone who lacks English proficiency will do less well in this test, and children were correctly excluded from analysis if their parents lacked English proficiency. Overall, the study children had below-average Stanford-Binet scores (89.8 ± 11.4). However, the standard method for calculating the composite score excludes subtests with a raw score of zero, and thus overestimates IQ in those children achieving a zero in any subtest. The Stanford-Binet IV score at 3 years of age does not correlate well with Wechsler Preschool and Primary Scale of Intelligence (WPPSI) scores at 4-5 years of age, but correlation is significantly improved by considering the number of subtests the child did not perform at 3 years of age (Grunau et al. 2000). The power of this study would therefore have been increased if the children had been assessed using the WPPSI test or if the authors had considered the number of zero-scored subtests in their analysis. However, if it is assumed that children in the Canfield study achieving a zero in any subtest are those with below-average IQs, the overall effect would have been to introduce a differential error in the estimates of IQ, that is, overestimating the IQ scores of children with higher lead levels. This would have biased (toward the null) the estimate of the slope of the relationship between blood lead and IQ and would have reduced the nonlinearity observed in Figure 1.

In correspondence after publication of the Canfield study, Bellinger and Needleman (2003) reanalyzed data from their prospective Boston cohort study, focusing on 48 children whose blood lead levels never exceeded 10 µg/dL at birth, 6, 12, 18, 24, 57, or 120 months. The regression coefficient was greater (-1.56) than that derived from analyses of children with peak blood lead levels > 10 µg/dL (-0.58), that is, their results replicated those of Canfield and colleagues. This reproducibility is of particular interest because the Boston cohort (high SES, average IQ of 105 at 2-4 years of age, mean blood lead 6.5 µg/dL at 2 years of age) was in many respects very different from the Rochester cohort (low SES, average IQ of 90 at 3 years of age, mean blood lead 9.7 µg/dL at 2 years of age). The authors conclude that residual confounding probably accounts for at least some of the disparity between the regression coefficients above and below 10 µg/dL, and because of this "the precise shape of the dose-effect relation at lead levels below 10 µg/dL remains uncertain" (Bellinger and Needleman 2003).

In summary, of the three most recent longitudinal studies that measured prenatal lead exposure (average blood lead ranging from 1 to 11.5 µg/dL), two found a negative association with subsequent IQ, and one found no effect. In contrast, all five recent longitudinal studies that measured postnatal exposure (average blood lead levels ranging from 6 to 44 µg/dL) found significant associations with cognitive development, and this association was maintained after adjusting for a range of covariates including child's sex and birth weight and parental/maternal IQ and years of education. The Port Pirie and Rochester studies considered the widest range of confounding factors and were the most robust methodologically. With the report of Canfield and colleagues and the recalculation of the Boston cohort results, these findings in nearly 1,300 children support an association between childhood lead exposure and subsequent cognitive impairment and extend the range of concern to children with lifetime average blood lead levels < 10 µg/dL.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.